arrowmodel

Dict-free, single-step conversion from Apache Arrow buffers to Pydantic v2 model instances.

arrowmodel uses a Rust core (via PyO3) to walk Arrow RecordBatch and Table buffers directly, building Pydantic model instances in a tight loop with no intermediate Python dicts. It replaces the to_pylist() + Model(**row) two-step with a single call that is roughly 2x faster for flat schemas.

Full Documentation

Quick Start

Installation

pip install arrowmodel

No Rust toolchain needed -- pre-built wheels are provided.

First Use

import pyarrow as pa
from arrowmodel import ArrowModel


class User(ArrowModel):
    id: int
    name: str
    score: float


batch = pa.record_batch(
    {
        "id": [1, 2, 3],
        "name": ["Alice", "Bob", "Carol"],
        "score": [9.5, 8.0, 7.3],
    }
)

users = User.convert(batch)
# [User(id=1, name='Alice', score=9.5), ...]

Key Features

• Dict-free conversion: Arrow buffers go straight to model_construct calls in Rust -- no to_pylist(), no intermediate dicts.
• ~2x faster for flat schemas: Roughly twice the throughput of the pure-Python approach, with less allocation pressure.
• Three API styles: ArrowModel base class for concise models, ArrowModelConverter for reusable converters, and model_convert/model_iter for one-shot use.
• Validated mode: Pass validate=True to run full Pydantic validation (model_validate_json) when you need type coercion or custom validators.
• Pydantic alias support: Resolves validation_alias, alias, field name, and populate_by_name / validate_by_name.
• Nested models: Arrow Struct columns map to nested Pydantic models, including optional structs and deeply nested hierarchies.
• Lazy iteration: iter() yields one model at a time, materialising one RecordBatch worth of instances in memory.
• Broad type coverage: Integers, floats, decimals, booleans, strings, dates, timestamps, durations, times, binary, lists, structs, maps, dictionaries, unions, intervals, and more.

Input Compatibility

arrowmodel accepts any Arrow-PyCapsule-compatible input via the Arrow C Data Interface:

• pyarrow -- RecordBatch, Table
• Polars -- DataFrame (exports via PyCapsule)
• nanoarrow -- arrays and record batches

No pyarrow dependency is required at runtime. If your Arrow data comes from Polars or nanoarrow, it works without installing pyarrow.

License

Apache-2.0